SDS 588: Artificial General Intelligence is Not Nigh

Brain Complexity

The complexity of biological brains far surpasses that of AI systems, a point emphasized by Jon Krohn. He argues that while AI models are increasing in parameters, they lack the anatomical intricacies of human brains. Biological brains, even those of simpler organisms, possess specialized structures like the amygdala and hippocampus, which are not yet fully understood or replicated in AI.

Modern AI systems make heavy use of the transformer architecture and then replicate that transformer architecture over and over to remarkable effect, no question. But this is meager complexity relative to the dozens of varieties of interacting anatomical components in the human brain.

--- Jon Krohn

Moreover, AI systems do not match the cellular complexity of biological brains, which include diverse cells like glial cells that contribute to intelligence 1 2.

   

Computational Barriers

Current computational models face significant barriers in replicating human-like intelligence. Jon Krohn notes that biological brains perform massively parallel processing, a feat that current computer systems cannot replicate. He suggests that achieving such parallel processing might require breakthroughs in quantum computing, which are not yet on the horizon.

While it might be only a decade or so until there is an algorithm with as many model parameters as there are connections between neurons and a human brain, it would take a vastly different computing approach than the one that prevails today.

--- Jon Krohn

The limitations of current technology highlight the challenges in achieving AGI, making the hypothesis that model scale alone will lead to AGI seem less credible 2 3.